ISR10 Scientific Report No. ISR-10 Information Storage and Retrieval The Indexing Function chapter Joseph John Rocchio Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. 2-17 to be relevant, the weights 0£ all terms contained in T are promoted C in the index image d.; while i£ anegative relevance judgment occurs, 1 the corresponding term weights are decreased. Over a number 0£ such operations those terms mbst use£ul in characterizing the content 0£ d6cume[OCRerr]t d. will receive increased signi£icance while terms 1 representing out 0£ context encodings should receive no net signi£icance gain. The e££iciency 0£ such a `scheme i's op'en[OCRerr]to [OCRerr]uestion. *C[OCRerr]early,' oneiwouldywant[OCRerr][OCRerr]the changes[OCRerr]in the index image 0£ a given document to re£lect some signi£icant sample 0£ user value judgments, since each user brings his own individual biases and viewpoints into the picture. The rate 0£ adjustment to the index images 0£ documents £or a given topic will then be in£luenced by the `density 6£ search re[OCRerr]uests £or *`[OCRerr]` that topic. This implies that' the documents 0£ most interest have the highest probability 0£ obtaining an improved index representation, a £eature which could be desirable. On the' other hand, however, a serious drawback 0£ this process lies in the di££iculty 0£ improving a poorly indexed document. Such a document will hardly ever be detected as being relevant to any user[OCRerr][OCRerr]s se[OCRerr]rch' re[OCRerr]uest so that any evaluations bbtained will be negative ones.' While spe'cial techniques (such as calling £or manual intervention) might be employed in this case, it se'ems clear tha't the value 0£ such an optimizationprocess will depend critically on ho'w good' the initial `index trans£ormation is to begin with. The question 0£ whether e££ective operational procedures can